Clean up build

build_tools/setup_helpers/extension.py

@@ -61,29 +61,31 @@ def _get_ela(debug):
 
 
 def _get_srcs():
-    return [str(p) for p in _CSRC_DIR.glob('**/*.cpp')]
+    srcs = [_CSRC_DIR / 'pybind.cpp']
+    srcs += list(_CSRC_DIR.glob('sox/**/*.cpp'))
+    if _BUILD_TRANSDUCER:
+        srcs += [_CSRC_DIR / 'transducer.cpp']
+    return [str(path) for path in srcs]
 
 
 def _get_include_dirs():
     dirs = [
         str(_ROOT_DIR),
     ]
-    if _BUILD_SOX:
+    if _BUILD_SOX or _BUILD_TRANSDUCER:
         dirs.append(str(_TP_INSTALL_DIR / 'include'))
-    if _BUILD_TRANSDUCER:
-        dirs.append(str(_TP_BASE_DIR / 'transducer' / 'submodule' / 'include'))
     return dirs
 
 
 def _get_extra_objects():
-    objs = []
+    libs = []
     if _BUILD_SOX:
         # NOTE: The order of the library listed bellow matters.
         #
         # (the most important thing is that dependencies come after a library
         # e.g., sox comes first, flac/vorbis comes before ogg, and
         # vorbisenc/vorbisfile comes before vorbis
-        libs = [
+        libs += [
             'libsox.a',
             'libmad.a',
             'libFLAC.a',
@@ -97,10 +99,12 @@ def _get_extra_objects():
             'libopencore-amrnb.a',
             'libopencore-amrwb.a',
         ]
-        for lib in libs:
-            objs.append(str(_TP_INSTALL_DIR / 'lib' / lib))
     if _BUILD_TRANSDUCER:
-        objs.append(str(_TP_BASE_DIR / 'build' / 'transducer' / 'libwarprnnt.a'))
+        libs += ['libwarprnnt.a']
+
+    objs = []
+    for lib in libs:
+        objs.append(str(_TP_INSTALL_DIR / 'lib' / lib))
     return objs
 
 
@@ -112,12 +116,20 @@ def _build_third_party():
     build_dir = str(_TP_BASE_DIR / 'build')
     os.makedirs(build_dir, exist_ok=True)
     subprocess.run(
-        args=['cmake', '..'],
+        args=[
+            'cmake',
+            f'-DCMAKE_INSTALL_PREFIX={_TP_INSTALL_DIR}',
+            f'-DBUILD_SOX={"ON" if _BUILD_SOX else "OFF"}',
+            f'-DBUILD_TRANSDUCER={"ON" if _BUILD_TRANSDUCER else "OFF"}',
+            '..'],
         cwd=build_dir,
         check=True,
     )
+    command = ['cmake', '--build', '.']
+    if _BUILD_TRANSDUCER:
+        command += ['--target', 'install']
     subprocess.run(
-        args=['cmake', '--build', '.'],
+        args=command,
         cwd=build_dir,
         check=True,
     )

third_party/CMakeLists.txt

@@ -1,9 +1,15 @@
-cmake_minimum_required(VERSION 3.1)
+cmake_minimum_required(VERSION 3.5)
 
 project(torchaudio_third_parties)
-include(ExternalProject)
 
-set(INSTALL_DIR ${CMAKE_CURRENT_SOURCE_DIR}/install)
+option(BUILD_SOX "Build libsox statically")
+option(BUILD_TRANSDUCER "Build transducer")
+
+SET(CMAKE_POSITION_INDEPENDENT_CODE ON)
+
+if (BUILD_SOX)
+include(ExternalProject)
+set(INSTALL_DIR ${CMAKE_INSTALL_PREFIX})
 set(ARCHIVE_DIR ${CMAKE_CURRENT_SOURCE_DIR}/archives)
 set(COMMON_ARGS --quiet --disable-shared --enable-static --prefix=${INSTALL_DIR} --with-pic --disable-dependency-tracking --disable-debug --disable-examples --disable-doc)
 
@@ -88,5 +94,8 @@ ExternalProject_Add(libsox
   # See https://github.com/pytorch/audio/pull/1026
   CONFIGURE_COMMAND ${CMAKE_CURRENT_SOURCE_DIR}/build_codec_helper.sh ${CMAKE_CURRENT_SOURCE_DIR}/src/libsox/configure ${COMMON_ARGS} --with-lame --with-flac --with-mad --with-oggvorbis --without-alsa --without-coreaudio --without-png --without-oss --without-sndfile --with-opus --with-amrwb --with-amrnb --disable-openmp --without-sndio --without-pulseaudio
 )
+endif()
 
+if(BUILD_TRANSDUCER)
 add_subdirectory(transducer)
+endif()

third_party/transducer/CMakeLists.txt

@@ -1,20 +1,7 @@
-CMAKE_MINIMUM_REQUIRED(VERSION 3.5)
-
-PROJECT(rnnt_release)
-
-SET(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -O2")
-
-IF(APPLE)
-    ADD_DEFINITIONS(-DAPPLE)
-ENDIF()
-
-INCLUDE_DIRECTORIES(submodule/include)
-
-SET(CMAKE_POSITION_INDEPENDENT_CODE ON)
-
 ADD_DEFINITIONS(-DRNNT_DISABLE_OMP)
 
 IF(APPLE)
+    ADD_DEFINITIONS(-DAPPLE)
     EXEC_PROGRAM(uname ARGS -v  OUTPUT_VARIABLE DARWIN_VERSION)
     STRING(REGEX MATCH "[0-9]+" DARWIN_VERSION ${DARWIN_VERSION})
     MESSAGE(STATUS "DARWIN_VERSION=${DARWIN_VERSION}")
@@ -30,9 +17,11 @@ ELSE()
 ENDIF()
 
 ADD_LIBRARY(warprnnt STATIC submodule/src/rnnt_entrypoint.cpp)
+target_include_directories(warprnnt PUBLIC submodule/include)
+set_target_properties(warprnnt PROPERTIES PUBLIC_HEADER submodule/include/rnnt.h)
 
-INSTALL(TARGETS warprnnt
-        LIBRARY DESTINATION "lib"
-        ARCHIVE DESTINATION "lib")
 
-INSTALL(FILES submodule/include/rnnt.h DESTINATION "submodule/include")
+INSTALL(
+  TARGETS warprnnt
+  ARCHIVE DESTINATION "lib"
+  PUBLIC_HEADER DESTINATION "include")

torchaudio/csrc/register.cpp → torchaudio/csrc/sox/register.cpp

@@ -2,7 +2,7 @@
 #include <torchaudio/csrc/sox/io.h>
 #include <torchaudio/csrc/sox/utils.h>
 
-TORCH_LIBRARY(torchaudio, m) {
+TORCH_LIBRARY_FRAGMENT(torchaudio, m) {
   //////////////////////////////////////////////////////////////////////////////
   // sox_utils.h
   //////////////////////////////////////////////////////////////////////////////
@@ -74,18 +74,4 @@ TORCH_LIBRARY(torchaudio, m) {
   m.def(
       "torchaudio::sox_effects_apply_effects_file",
       &torchaudio::sox_effects::apply_effects_file);
-
-  //////////////////////////////////////////////////////////////////////////////
-  // transducer.cpp
-  //////////////////////////////////////////////////////////////////////////////
-  #ifdef BUILD_TRANSDUCER
-  m.def("rnnt_loss(Tensor acts,"
-                  "Tensor labels,"
-                  "Tensor input_lengths,"
-                  "Tensor label_lengths,"
-                  "Tensor costs,"
-                  "Tensor grads,"
-                  "int blank_label,"
-                  "int num_threads) -> int");
-  #endif
 }

torchaudio/csrc/transducer.cpp

@@ -1,5 +1,3 @@
-#ifdef BUILD_TRANSDUCER
-
 #include <iostream>
 #include <numeric>
 #include <string>
@@ -76,7 +74,16 @@ int64_t cpu_rnnt_loss(torch::Tensor acts,
 }
 
 TORCH_LIBRARY_IMPL(torchaudio, CPU, m) {
-    m.impl("rnnt_loss", &cpu_rnnt_loss);
+  m.impl("rnnt_loss", &cpu_rnnt_loss);
 }
 
-#endif
+TORCH_LIBRARY_FRAGMENT(torchaudio, m) {
+  m.def("rnnt_loss(Tensor acts,"
+                  "Tensor labels,"
+                  "Tensor input_lengths,"
+                  "Tensor label_lengths,"
+                  "Tensor costs,"
+                  "Tensor grads,"
+                  "int blank_label,"
+                  "int num_threads) -> int");
+}